Voice and dictation

Talk instead of typing. Audio is transcribed on the device.

Press the microphone in the composer and start speaking. The button turns green while it is listening. Text appears in the box as you talk — the agent transcribes each phrase when you pause, so you do not have to stop and wait at the end.

The chat composer while dictation is active: the microphone button is green and pulsing. — Red means the microphone is off; green means it is listening. The same green as the Online dot.

Voice replies

Dictation is you talking to the agent; voice mode is the agent talking back. Type /voice on in chat to have replies spoken aloud, /voice off to silence them, and /voice status to see the current setting. The first spoken reply after a restart takes longer while the speech model loads.

Where the audio goes

Nowhere. Speech is transcribed by a model running on the device itself, so dictation works without an internet connection and your voice is never uploaded.

If nothing appears

Check the browser or app has microphone permission.
Speech-to-text must be enabled — Config → stt → enabled.
The first use after a restart is slower, because the transcription model has to load.